nb_kernel333_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,082 行 · 第 1/5 页
S
2,082 行
movaps nb333_qqM(%esp),%xmm0 addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV mulps %xmm0,%xmm5 ## vcoul=qq*VV mulps %xmm7,%xmm0 ## fijC=FF*qq ## at this point mm5 contains vcoul and xmm0 fijC ## increment vcoul - then we can get rid of mm5 addps nb333_vctot(%esp),%xmm5 movaps %xmm5,nb333_vctot(%esp) addps nb333_rinvO(%esp),%xmm0 ## total fscal (temp. storage in rinvO) xorps %xmm4,%xmm4 mulps nb333_rinvM(%esp),%xmm0 mulps nb333_tsc(%esp),%xmm0 subps %xmm0,%xmm4 movaps nb333_dxO(%esp),%xmm0 movaps nb333_dyO(%esp),%xmm1 movaps nb333_dzO(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 now contains tx-tz (partial force) movss nb333_fixO(%esp),%xmm3 movss nb333_fiyO(%esp),%xmm4 movss nb333_fizO(%esp),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,nb333_fixO(%esp) movss %xmm4,nb333_fiyO(%esp) movss %xmm5,nb333_fizO(%esp) ## updated the O force now do the H's movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $0x39,%xmm3,%xmm3 ## shift right shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb333_fixH1(%esp),%xmm3 addss nb333_fiyH1(%esp),%xmm4 addss nb333_fizH1(%esp),%xmm5 movss %xmm3,nb333_fixH1(%esp) movss %xmm4,nb333_fiyH1(%esp) movss %xmm5,nb333_fizH1(%esp) ## updated the H1 force shufps $0x39,%xmm3,%xmm3 shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb333_fixH2(%esp),%xmm3 addss nb333_fiyH2(%esp),%xmm4 addss nb333_fizH2(%esp),%xmm5 movss %xmm3,nb333_fixH2(%esp) movss %xmm4,nb333_fiyH2(%esp) movss %xmm5,nb333_fizH2(%esp) ## updated the H2 force movl nb333_faction(%ebp),%edi shufps $0x39,%xmm3,%xmm3 shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb333_fixM(%esp),%xmm3 addss nb333_fiyM(%esp),%xmm4 addss nb333_fizM(%esp),%xmm5 movss %xmm3,nb333_fixM(%esp) movss %xmm4,nb333_fiyM(%esp) movss %xmm5,nb333_fizM(%esp) ## updated the M force movd %mm0,%eax ## the fj's - move in from mem start by acc. tx/ty/tz in xmm0, xmm1 movlps (%edi,%eax,4),%xmm6 movss 8(%edi,%eax,4),%xmm7 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,%xmm0 movaps %xmm4,%xmm1 movaps %xmm5,%xmm2 shufps $0x39,%xmm3,%xmm3 ## shift right shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 unpcklps %xmm1,%xmm0 ## x,y sum in xmm0, z sum in xmm2 subps %xmm0,%xmm6 subss %xmm2,%xmm7 movlps %xmm6,(%edi,%eax,4) movss %xmm7,8(%edi,%eax,4) decl nb333_innerk(%esp) jz _nb_kernel333_ia32_sse.nb333_updateouterdata jmp _nb_kernel333_ia32_sse.nb333_odd_loop_nb_kernel333_ia32_sse.nb333_updateouterdata: movl nb333_ii3(%esp),%ecx movl nb333_faction(%ebp),%edi movl nb333_fshift(%ebp),%esi movl nb333_is3(%esp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb333_fixO(%esp),%xmm0 movaps nb333_fiyO(%esp),%xmm1 movaps nb333_fizO(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss (%edi,%ecx,4),%xmm3 movss 4(%edi,%ecx,4),%xmm4 movss 8(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,(%edi,%ecx,4) movss %xmm4,4(%edi,%ecx,4) movss %xmm5,8(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift movaps %xmm0,%xmm6 movss %xmm2,%xmm7 movlhps %xmm1,%xmm6 shufps $8,%xmm6,%xmm6 ## constant 00001000 ## accumulate H1i forces in xmm0, xmm1, xmm2 movaps nb333_fixH1(%esp),%xmm0 movaps nb333_fiyH1(%esp),%xmm1 movaps nb333_fizH1(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss 12(%edi,%ecx,4),%xmm3 movss 16(%edi,%ecx,4),%xmm4 movss 20(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,12(%edi,%ecx,4) movss %xmm4,16(%edi,%ecx,4) movss %xmm5,20(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movaps nb333_fixH2(%esp),%xmm0 movaps nb333_fiyH2(%esp),%xmm1 movaps nb333_fizH2(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss 24(%edi,%ecx,4),%xmm3 movss 28(%edi,%ecx,4),%xmm4 movss 32(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,24(%edi,%ecx,4) movss %xmm4,28(%edi,%ecx,4) movss %xmm5,32(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## accumulate Mi forces in xmm0, xmm1, xmm2 movaps nb333_fixM(%esp),%xmm0 movaps nb333_fiyM(%esp),%xmm1 movaps nb333_fizM(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss 36(%edi,%ecx,4),%xmm3 movss 40(%edi,%ecx,4),%xmm4 movss 44(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,36(%edi,%ecx,4) movss %xmm4,40(%edi,%ecx,4) movss %xmm5,44(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## increment fshift force movlps (%esi,%edx,4),%xmm3 movss 8(%esi,%edx,4),%xmm4 addps %xmm6,%xmm3 addss %xmm7,%xmm4 movlps %xmm3,(%esi,%edx,4) movss %xmm4,8(%esi,%edx,4) ## get n from stack movl nb333_n(%esp),%esi ## get group index for i particle movl nb333_gid(%ebp),%edx ## base of gid[] movl (%edx,%esi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb333_vctot(%esp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addps %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now movaps %xmm7,%xmm6 shufps $1,%xmm6,%xmm6 addss %xmm6,%xmm7 ## add earlier value from mem movl nb333_Vc(%ebp),%eax addss (%eax,%edx,4),%xmm7 ## move back to mem movss %xmm7,(%eax,%edx,4) ## accumulate total lj energy and update it movaps nb333_Vvdwtot(%esp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addps %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now movaps %xmm7,%xmm6 shufps $1,%xmm6,%xmm6 addss %xmm6,%xmm7 ## add earlier value from mem movl nb333_Vvdw(%ebp),%eax addss (%eax,%edx,4),%xmm7 ## move back to mem movss %xmm7,(%eax,%edx,4) ## finish if last movl nb333_nn1(%esp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel333_ia32_sse.nb333_outerend ## not last, iterate outer loop once more! movl %esi,nb333_n(%esp) jmp _nb_kernel333_ia32_sse.nb333_outer_nb_kernel333_ia32_sse.nb333_outerend: ## check if more outer neighborlists remain movl nb333_nri(%esp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel333_ia32_sse.nb333_end ## non-zero, do one more workunit jmp _nb_kernel333_ia32_sse.nb333_threadloop_nb_kernel333_ia32_sse.nb333_end: emms movl nb333_nouter(%esp),%eax movl nb333_ninner(%esp),%ebx movl nb333_outeriter(%ebp),%ecx movl nb333_inneriter(%ebp),%edx movl %eax,(%ecx) movl %ebx,(%edx) movl nb333_salign(%esp),%eax addl %eax,%esp addl $988,%esp popl %edi popl %esi popl %edx popl %ecx popl %ebx popl %eax leave ret.globl nb_kernel333nf_ia32_sse.globl _nb_kernel333nf_ia32_ssenb_kernel333nf_ia32_sse: _nb_kernel333nf_ia32_sse: .set nb333nf_p_nri, 8.set nb333nf_iinr, 12.set nb333nf_jindex, 16.set nb333nf_jjnr, 20.set nb333nf_shift, 24.set nb333nf_shiftvec, 28.set nb333nf_fshift, 32.set nb333nf_gid, 36.set nb333nf_pos, 40.set nb333nf_faction, 44.set nb333nf_charge, 48.set nb333nf_p_facel, 52.set nb333nf_argkrf, 56.set nb333nf_argcrf, 60.set nb333nf_Vc, 64.set nb333nf_type, 68.set nb333nf_p_ntype, 72.set nb333nf_vdwparam, 76.set nb333nf_Vvdw, 80.set nb333nf_p_tabscale, 84.set nb333nf_VFtab, 88.set nb333nf_invsqrta, 92.set nb333nf_dvda, 96.set nb333nf_p_gbtabscale, 100.set nb333nf_GBtab, 104.set nb333nf_p_nthreads, 108.set nb333nf_count, 112.set nb333nf_mtx, 116.set nb333nf_outeriter, 120.set nb333nf_inneriter, 124.set nb333nf_work, 128 ## stack offsets for local variables ## bottom of stack is cache-aligned for sse use .set nb333nf_ixO, 0.set nb333nf_iyO, 16.set nb333nf_izO, 32.set nb333nf_ixH1, 48.set nb333nf_iyH1, 64.set nb333nf_izH1, 80.set nb333nf_ixH2, 96.set nb333nf_iyH2, 112.set nb333nf_izH2, 128.set nb333nf_ixM, 144.set nb333nf_iyM, 160.set nb333nf_izM, 176.set nb333nf_iqM, 192.set nb333nf_iqH, 208.set nb333nf_qqM, 224.set nb333nf_qqH, 240.set nb333nf_rinvO, 256.set nb333nf_rinvH1, 272.set nb333nf_rinvH2, 288.set nb333nf_rinvM, 304.set nb333nf_rO, 320.set nb333nf_rH1, 336.set nb333nf_rH2, 352.set nb333nf_rM, 368.set nb333nf_tsc, 384.set nb333nf_c6, 400⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?